See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/321496935

I believe, therefore I achieve (and vice versa): A meta-analytic cross-

lagged panel analysis of self-efficacy and academic performance

Article  in  Learning and Individual Differences · January 2018

DOI: 10.1016/j.lindif.2017.11.015

CITATIONS READS

0 563

4 authors:

Kate Talsma Benjamin Schüz

University of Tasmania Universität Bremen
1 PUBLICATION   0 CITATIONS    106 PUBLICATIONS   2,668 CITATIONS   

SEE PROFILE SEE PROFILE

Ralf Schwarzer Kimberley Norris

Freie Universität Berlin University of Tasmania
413 PUBLICATIONS   23,056 CITATIONS    25 PUBLICATIONS   84 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Academic achievement behaviours View project

Adaptation and resilience in extreme environments View project

All content following this page was uploaded by Kimberley Norris on 04 December 2017.

The user has requested enhancement of the downloaded file.

 Learning and Individual Differences 61 (2018) 136–150

Contents lists available at ScienceDirect

Learning and Individual Differences

journal homepage: www.elsevier.com/locate/lindif

I believe, therefore I achieve (and vice versa): A meta-analytic cross-lagged T

panel analysis of self-efficacy and academic performance
⁎
Kate Talsmaa, , Benjamin Schüza,b, Ralf Schwarzerc,d, Kimberley Norrisa
a
University of Tasmania, School of Medicine (Psychology), Tasmania, Australia
b
Universität Bremen, Department of Human and Health Sciences, Bremen, Germany
c
Freie Universität Berlin, Department of Psychology, Berlin, Germany
d
SWPS University of Social Sciences and Humanities, Department of Clinical, Health, and Rehabilitation Psychology, Wroclaw, Poland

A R T I C L E I N F O A B S T R A C T

Keywords: Self-efficacy has long been viewed as an important determinant of academic performance. A counter-position is
Self-efficacy that self-efficacy is merely a reflection of past performance. Research in the area is limited by unidirectional
Academic performance designs which cannot address reciprocity or the comparative strength of directional effects. This systematic
Reciprocal effects review and meta-analysis considered both directions of the relationship simultaneously, pooling data from
Cross-lagged panel analysis
longitudinal studies measuring both academic self-efficacy and academic performance over two waves. Pooled
Meta-analysis
correlations (k = 11, N = 2688) were subjected to cross-lagged path analysis that provided support for a re-
ciprocal effects model. Performance had a net positive effect on subsequent self-efficacy (β = 0.205,
p < 0.001), significantly larger than the effect of self-efficacy on performance (β = 0.071, p < 0.001).
Moderator analyses indicated that reciprocity holds for adults, but not for children (in whom performance
uniquely impacts subsequent self-efficacy beliefs, but not the reverse). Cross-lagged effects were stronger in
studies which used methodologies consistent with recommendations of self-efficacy theorists.

1. Introduction 1.1. Self-efficacy → academic performance (I believe; therefore I achieve)

The non-intellective antecedents of student performance are of
great interest to educators and education researchers (Robbins et al.,
2004; Stankov & Lee, 2014), and research in this area is an important
determinant of education policy (Bong, 2012; Pajares & Usher,
2008). One construct which has received a great deal of research
attention is perceived self-efficacy – a core dimension of human
agency widely believed to be positively related to academic success
(Bandura, 1977, 1997; Bong, 2012; Pajares & Schunk, 2001). Self-
efficacy refers to an individual's perception of their own capability to
organise and execute required courses of action to achieve particular
outcomes (Bandura, 1977, 1997). Self-efficacy is believed to enhance
performance through a range of mechanisms: individuals with high
levels of self-efficacy set more difficult goals, expend more effort,
persist for longer with challenges, and show resilience in the face of
adversity (Klassen & Usher, 2010). These achievements in turn are
assumed to increase self-efficacy, which results in a self-fulfilling
prophecy process (Bandura, 1977, 1997).
A vast body of research has explored the idea that self-efficacy is the
antecedent in the relationship and exerts a positive motivational in-
fluence on performance (Honicke & Broadbent, 2016; Vancouver,
Thompson, & Williams, 2001). Such research takes its lead from early
studies by Bandura and colleagues (see Pajares, 1997; Zimmerman,
2000 for review), which demonstrated that self-efficacy influenced
subsequent behaviour. Self-efficacy → performance research also draws
on the definition of self-efficacy as a future-oriented, predictive con-
struct: measures of self-efficacy involve statements of confidence in
ability to achieve a future performance goal (Bong, 2012). From an
applied perspective, research on this relationship draws impetus from
(and feeds into) educational settings, in which interventions are sought
to improve performance.
In a review of self-efficacy → performance research conducted over
the past three decades, Klassen and Usher (2010) describe self-efficacy
as a crucial and powerful influence on academic performance, ac-
counting for approximately a quarter of the variance in outcomes. Self-

⁎
Corresponding author at: University of Tasmania, School of Medicine, Psychology, Private Bag 30, Hobart, TAS 7001, Australia.
E-mail address: kate.talsma@utas.edu.au (K. Talsma).

https://doi.org/10.1016/j.lindif.2017.11.015
Received 6 April 2017; Received in revised form 9 October 2017; Accepted 22 November 2017
1041-6080/ © 2017 Elsevier Inc. All rights reserved.
K. Talsma et al.
efficacy is argued to rival previous performance and mental ability in its
power to predict academic performance (Pajares & Kranzler, 1995).
Meta-analyses of cross-sectional studies consistently point to self-effi-
cacy as one of the strongest correlates of academic performance.
Richardson, Abraham, and Bond (2012) found that self-efficacy was the
strongest correlate of tertiary GPA (ρ = 0.59), exceeding high school
GPA (ρ = 0.41), scholastic aptitude tests (ρ = 0.31–0.33), and in-
telligence (ρ = 0.21). Similar relationships are reported by Multon,
Brown, and Lent (1991) and Honicke and Broadbent (2016).
Longitudinal research in the self-efficacy → academic performance
paradigm is comparatively sparse (Honicke & Broadbent, 2016).
Nonetheless, findings are consistent across school-aged and tertiary
samples, in that self-efficacy positively predicts subsequent academic
performance, over periods ranging from a single semester, to courses
over several years (rs = 0.37–0.52; Chiang & Lin, 2014; Garriott &
Flores, 2013; Majer, 2009; Parker, Marsh, Ciarrochi, Marshall, &
Abduljabbar, 2014; Phan & Ngu, 2016).
Based on such findings, direct manipulation of self-efficacy has long
been recommended as an intervention strategy in learning settings
(e.g., Bong, 2012; Pajares & Usher, 2008; Zimmerman & Bandura,
1995). Given the practical implications for educational reforms and
interventions, as well as the implications for theory and research, it is
important to be confident that the relationship between self-efficacy
and academic performance is being interpreted accurately (Valentine,
DuBois, & Cooper, 2004).
1.2. Academic performance → self-efficacy (I achieve; therefore I believe)
As the bulk of research on the relationship between self-efficacy and
performance is cross-sectional, inferences about the direction of influ-
ence are impossible (Honicke & Broadbent, 2016; Pajares & Usher,
2008). Some researchers argue that cross-sectional relationships reflect
the influence of performance on subsequent self-efficacy, not the re-
verse (Shea & Howell, 2000; Vancouver et al., 2001). Heggestad and
Kanfer (2005) argue that self-efficacy is simply a proxy for past per-
formance with no unique predictive power.
There is little doubt that performance outcomes influence self-effi-
cacy. Previous mathematics performance positively predicts mathe-
matics self-efficacy in both school-aged and tertiary samples (e.g.,
Klassen, 2004; Matsui, Matsui, & Ohnishi, 1990). Chin and Kameoka
(2002) reported that reading scores predicted the educational self-ef-
ficacy of high school students (β = 0.32) after accounting for a range of
demographic and psychosocial predictors. In a recent study, both
standardised test scores and first semester GPA correlated with self-ef-
ficacy 12 months later, r = 0.30 (Lee, Flores, Navarro, & Kanagui-
Munoz, 2015).
Such findings are not inconsistent with self-efficacy theory; in fact,
investigations of the academic performance → self-efficacy relationship
often grow out of self-efficacy theory directly. Mastery experience (an
individual's experience of performance success) is one of four posited
sources of self-efficacy beliefs, along with vicarious experience, verbal
social persuasion, and emotional physiological arousal (Bandura,
1997). Research shows that, of these four, mastery experience is the
strongest (if not the only) predictor of self-efficacy (Britner & Pajares,
2006; Lent, Lopez, & Bieschke, 1991; Usher & Pajares, 2008).
While self-efficacy theorists do not discount the influence of per-
formance on self-efficacy, they refute the argument that self-efficacy
has no unique impact on performance (Bandura, 2012; cf. Heggestad &
Kanfer, 2005). Some researchers have attempted to rule out the hy-
pothesis that self-efficacy provides no incremental prediction of per-
formance beyond that accounted for by previous performance, by meta-
analysing the self-efficacy → performance relationship, controlling for
previous performance. Valentine et al. (2004) used this approach, in-
terpreting the unique self-efficacy → performance effect (β = 0.10,
k = 9) as “small but noteworthy” (p. 127). Robbins et al. (2004) found
that the prediction of academic achievement by academic self-efficacy
137
Learning and Individual Differences 61 (2018) 136–150
was incremental to that of socioeconomic status, standardised
achievement measures, and high school GPA (β = 0.20, k = 18).
The research summarised above provides evidence of both a self-
efficacy → performance relationship and a performance → self-efficacy
relationship; it also suggests that self-efficacy has an effect on sub-
sequent performance incremental to that of previous performance
alone. While this longitudinal research extends considerably on pre-
vious cross-sectional findings, it does not address the possibility that
self-efficacy and performance are reciprocally related – nor does it
elucidate the relative strength of directional effects.
1.3. A chicken-and-egg conundrum
The question of the direction of causality in the relationship be-
tween self-beliefs and academic performance has been described as one
of “thorniest issues” in this area (Pajares & Schunk, 2001). According to
social cognitive theory, self-efficacy exists in a framework of reciprocal
determinism: behaviour both shapes, and is shaped by a range of in-
teracting factors (Bandura, 1977, 1997). In this model, self-efficacy and
performance modify each other iteratively within a constant feedback
loop (Multon et al., 1991). In educational settings, learners reflect on
their performance and use this information when formulating their self-
efficacy beliefs, which then influence subsequent performance (Phan,
2012).
Several recent studies demonstrate increased interest in the mutual
influences of self-efficacy and academic performance over time. In
longitudinal studies of both high school and university students in
which multiple measurements of self-efficacy and performance are
staggered over several years, self-efficacy and performance predict each
other, either in a self-efficacy → performance → self-efficacy pattern, or
in a performance → self-efficacy → performance pattern (e.g., Caprara,
Vecchione, Alessandri, Gerbino, & Barbaranelli, 2011; Hwang, Choi,
Lee, Culver, & Hutchison, 2016). These types of designs provide evi-
dence of positive mutual temporal effects, but by staggering data col-
lection over time they do not enable the modelling of simultaneous
reciprocal effects (Rogosa, 1988).
Thus, while reciprocal determinism between academic self-efficacy
and academic performance may be considered a fait accompli
(Lindsley, Brass, & Thomas, 1995), there is little direct empirical evi-
dence which supports this proposition (Williams & Williams, 2010).
This gap in the literature is likely due, in part, to the paucity of long-
itudinal studies in the area, a problem which is compounded because
the two unidirectional research paradigms in this area are pursued
largely independently of each other (Shea & Howell, 2000). In the most
recent self-efficacy → performance meta-analysis, Honicke and
Broadbent (2016) suggest a reciprocal relationship exists and re-
commend that this be investigated directly.
In the case of self-efficacy and performance, the issue of how to
assess the unique influence of one variable on the other has been
characterised by debate. Controlling for raw past performance (e.g.,
Feltz, 1982; Mitchell, Hopper, Daniels, George-Falvy, & James, 1994) is
argued to be an over-correction (Bandura & Locke, 2003), while re-
sidualising self-efficacy from past performance (Bandura & Locke,
2003) is argued to lead to statistically artefactual results (Heggestad &
Kanfer, 2005). In fact, as both past self-efficacy and past performance
are expected to be covarying common-cause variables (Bandura, 2012;
Feltz, Chow, & Hepler, 2008; Heggestad & Kanfer, 2005; Vancouver
et al., 2001), any unidirectional approach will result in inflated esti-
mates of the influence of these variables on each other (Brown et al.,
2008). An approach is required where both self-efficacy and perfor-
mance can be controlled at time 1 (Bandura, 2012; Feltz et al., 2008).
One approach that may overcome these limitations is cross-lagged
panel analysis (CLPA; see Fig. 1); a uniquely powerful approach to
chicken-and-egg questions (Singh & Tyagi, 2014) which has been
gaining traction in the behavioral sciences literature (e.g., Riketta,
2008). CLPA provides more robust evidence of potential reciprocal
K. Talsma et al.
causality within dynamic relationships than other models, by meeting
the requirement for temporal precedence, and demonstrating the un-
ique effect of variable X on variable Y, and vice versa, controlling for
other model paths (Selig & Little, 2012).
1.4. Main aims of the present study
This study aims to explore the reciprocity and comparative strength
of net directional effects in the self-efficacy/academic performance re-
lationship by meta-analysing longitudinal panel studies (two-variable,
two-wave designs). Pooled correlations will be fit to a CLPA model
(Fig. 1) comprising two auto-correlations/stability coefficients (βSE1SE2
and βP1P2), two synchronous/cross-sectional correlations (βSE1P1 and
βSE2P2), and two cross-lagged correlations (βSE1P2 and βP1SE2). A re-
ciprocal relationship is demonstrated by significant positive cross-
lagged paths; the stronger antecedent, if any, is marked by a larger
coefficient.
A further aim of the present review is to examine variables which
moderate the self-efficacy ↔ performance relationship. Moderator
analyses are critical given the consistent finding of substantial hetero-
geneity in the self-efficacy/performance relationship (Honicke &
Broadbent, 2016; Multon et al., 1991; Richardson et al., 2012;
Valentine et al., 2004). Further to accounting for heterogeneity, these
analyses contribute to our understanding of the conditions under which
self-efficacy and academic performance are most strongly related –
taking previous findings within a unidirectional self-efficacy → perfor-
mance paradigm and extending these to the reciprocal self-efficacy ↔
performance relationship framework.
In summary, the main aims of this study are: to conduct a meta-
analytic CLPA on studies presenting panel data regarding self-efficacy
and academic performance, to test whether the data are consistent with
reciprocal effects, to explore the relative antecedent strength of the two
variables, and to test whether the relationship varies as a function of
participant- and study-level variables identified as potential moderators
in the literature.
1.5. Moderators of the self-efficacy ↔ performance relationship
1.5.1. Age
Participant age has been found to lead to differences in self-efficacy
levels and variations in the relationship between self-efficacy and per-
formance (e.g., Davis-Kean et al., 2008). The earliest meta-analysis of
the relationship between self-efficacy and academic performance
(Multon et al., 1991) reported a stronger correlation for college stu-
dents (r = 0.35) than elementary students (r = 0.21). This difference
likely relates to different levels of cognitive maturity, with adults more
138
Learning and Individual Differences 61 (2018) 136–150
Fig. 1. Conceptual diagram of the cross-lagged path
model.
likely than children to use meta-cognitive strategies (Flavell, 1979) and
reflect on self-constructs (Harter, 1999). Therefore, it is anticipated that
the modelled relationships will be stronger for adults than for children.
1.5.2. Sex
Participant sex may also moderate the reciprocal relationship.
Theoretically, differences may grow out of self-fulfilling prophecies
whereby females are socialised to view themselves as less capable than
males (Williams & Williams, 2010). Characteristics of educational set-
tings may be implicated in this process, with associated differences in
competence perceptions and performance (Eccles & Blumenfeld, 1985).
Meta-analyses have reported sex differences in regard to academic
performance (Richardson et al., 2012) and self-efficacy (Huang, 2013).
However, Valentine et al. (2004) found that sex was not a moderator
when self-efficacy was used to predict academic performance. Given the
lack of consistency in findings, this moderator analysis will be ex-
ploratory in nature.
1.5.3. Self-efficacy scale type
The type of scale most appropriate for self-efficacy measurement is
the subject of debate, and it is anticipated that scale differences will
moderate findings. Bandura (2006) recommends the use of unipolar
scales, with participants providing ratings of their degree of confidence
to perform a task, from 0 to 100%. Unipolar scales are argued to be
more sensitive and reliable (Bandura, 1986). While the use of Likert
scales is criticised (Klassen & Usher, 2010; Lee & Bobko, 1994) it re-
mains very common (see Table 2). Likert scales are argued to be in-
appropriate for a variable which does not have a positive/negative
valence with a neutral midpoint (Bandura, 2012). Pajares, Hartley, and
Valiante (2001) compared unipolar and Likert scales and found uni-
polar measures to be more psychometrically sound. Thus, it is hy-
pothesised that models based on studies using unipolar scales will de-
monstrate stronger cross-lagged paths than those using Likert scales.
1.5.4. Lag between measurement waves
Variation in lag time between panel measurements is also con-
sidered as a moderating factor. In educational research, decisions about
measurement timing often lack a strong rationale, resting instead on
pragmatic considerations such as key academic dates (e.g., Phan,
2012), or conventions for ease of interpretation (e.g., Valentine et al.,
2004). However, shorter lags between measurement waves are likely to
be associated with stronger effects (Bandura, 1997; Gist & Mitchell,
1992; Moriarty, 2014), potentially owing to intervening variables
‘washing out’ the relationship over time (Honicke & Broadbent, 2016;
Valentine et al., 2004). Recent research findings appear to be consistent
with this position (Galyon, Blondin, Yaw, Nalls, & Williams, 2012; Gore
K. Talsma et al.
Jr, 2006; Obrentz, 2012; Zusho, Pintrich, & Coppola, 2003). Thus, we
hypothesise that shorter lags will be associated with stronger effects
than longer lags.
1.5.5. Specificity of measurement/match between specificity of self-efficacy
and performance
Measurement issues regarding the degree of specificity of the self-
efficacy measure and the degree of match between the self-efficacy and
performance measures have been discussed for many years (Bandura,
1986). Further, these issues have been demonstrated to be critical to the
accuracy of prediction of behaviour (Pajares & Usher, 2008). Domain-
based self-efficacy measures reflect confidence in one's ability to per-
form academically in a subject or particular course; e.g., I'm sure I can do
even the hardest work in my math class (Lewis et al., 2012). Domain-based
self-efficacy can also be measured by asking individuals to rate their
levels of confidence in achieving a range of grades on a subject-specific
course (e.g., Beck & Schmidt, 2013). Task-based self-efficacy measures
reflect confidence in one's ability to perform particular academic tasks;
often, this will be answering specific questions correctly (e.g., Bonne &
Johnston, 2016) or attaining a particular grade on an isolated perfor-
mance task such as an exam (e.g., Mone, Baker, & Jeffries, 1995). It is
generally reported that more specific self-efficacy measures are more
strongly correlated with performance than less specific measures, and
more closely matched measures of self-efficacy and performance are
more strongly correlated with each other (Bong, 2001; Pajares & Miller,
1995; Zimmerman & Bandura, 1995). In addition, reviewers report that
measurement specificity (Multon et al., 1991) and match (Valentine
et al., 2004) moderate the unidirectional relationship between self-ef-
ficacy and performance in meta-analyses. Thus, it is hypothesised that
models based on studies using more specific self-efficacy measures, and
matched specificity of self-efficacy and performance, will demonstrate
stronger model paths than those with less specific measures and non-
matched specificity.
1.5.6. Cohort effects
We note that, as no restrictions were put in place in terms of date of
article publication, it is possible that cohort effects may be found in the
sample. Broad changes in educational policies and assessment practices
over time may affect the self-efficacy ↔ performance relationship
(Goldstein, 1983). An exploratory moderator analysis using time since
publication as a variable will be undertaken to test this possibility.
1.6. Summary
By using meta-analytic cross-lagged panel analysis, this systematic
review addresses the chicken-and-egg question of self-efficacy and
performance, bringing together two avenues of enquiry which have so
far been largely isolated (self-efficacy → performance versus perfor-
mance → self-efficacy). To our knowledge, a meta-analytic cross-lagged
panel analysis is yet to be conducted with the constructs of self-efficacy
and academic performance.1 The present review builds on others (e.g.,
Richardson et al., 2012; Valentine et al., 2004) in several important
ways. Firstly, the present analysis assesses the net effect of self-efficacy
on performance while holding all CLPA model paths constant, rather
than just the effect of previous performance. Secondly, this is the first
analysis, to our knowledge, of the net effect of academic performance
on subsequent academic self-efficacy, accounting for other model paths
including initial levels of self-efficacy. Finally, our approach applies
stricter inclusion and exclusion criteria than previous reviews in order
1
It should be noted that, in a meta-analysis of cross-lagged panel studies of the broader
self-efficacy/performance relationship by Sitzmann and Yeo (2013), a subset of studies
were related to self-efficacy and academic performance. However, the methodological
rigour of this study and its overall contribution to the literature has been questioned
(Bandura, 2015). For example, it has been noted that some included studies did not
measure one of the two key constructs (Bandura, 2015).
139
Learning and Individual Differences 61 (2018) 136–150
to ensure adequate construct validity and obtain a sample of more
homogeneous studies that will allow testing the proposed relationships
with more reliability than previous studies.
A better understanding of the complexities of the relationship be-
tween self-efficacy and academic performance will provide valuable
input into theory development and research design, and will also help
to inform education policy and teacher training programs.
2. Method
This systematic review and meta-analysis was undertaken in ac-
cordance with the PRISMA guidelines (Moher, Liberati, Tetzlaff, &
Altman, 2009).
2.1. Inclusion criteria
2.1.1. Operationalisation of self-efficacy
There is evidence that measurement issues pose a genuine threat to
the validity of research in this area (Bandura, 2012; Klassen & Usher,
2010). Previous reviews report that differences in the operationalisa-
tion of both self-efficacy and performance account for heterogeneity
across studies (Honicke & Broadbent, 2016; Multon et al., 1991;
Richardson et al., 2012; Valentine et al., 2004). Self-efficacy is often not
appropriately measured, nor differentiated from other constructs such
as self-esteem and self-concept (Bandura, 2012; Klassen & Usher, 2010;
Lee & Bobko, 1994; Pajares, 1996). This problem is compounded when
reviews cast a wide net, because very broad definitions of variables may
result in the conglomeration of distinct constructs (Bandura, 2012). To
address these issues, it has been recommended that researchers use
more narrowly defined and theoretically grounded measures – which
we will apply in this present review (Klassen & Usher, 2010; Robbins
et al., 2004; Valentine et al., 2004). We include studies when scales
reflect the three key components of self-efficacy: a personal judgement
of ability (internal attribution) to prospectively (future orientation)
perform academically (a behaviour, as opposed to an attitude or per-
sonal characteristic) (Bandura, 2006; Schwarzer & McAuley, 2016).
Global/generalised/“trait” measures will not be included, owing to
theoretical concerns about their applicability and their limited power to
predict academic performance (Bandura, 2012; Klassen & Usher, 2010;
Pajares, 1997). Furthermore, this review does not include efficacy for
self-regulated learning, e.g., How well can you arrange a place to study
without distractions? (Zimmerman, Bandura, & Martinez-Pons, 1992). In
sum, we seek measures which reflect a can do judgement for future
academic performance (Table 1).
2.1.2. Operationalisation of academic performance
Previous reviews have also suffered from inconsistent oper-
ationalisations of academic performance (Robbins et al., 2004). Multon
et al. (1991) and Honicke and Broadbent (2016) used a wide range of
performance metrics in their reviews, and both reported that perfor-
mance specification moderated the relationships found. A more precise
definition of academic performance would contribute to the validity of
analyses. While GPA is a readily accessible measure of academic per-
formance, aggregation potentially blurs performance across subject
areas, task types, and – most importantly for the present analysis –
across time. To provide measurement precision, we define academic
performance as objective scores on individual performance tasks, such
as tests or exams. Such a definition is also consistent with the re-
commendations of self-efficacy theorists. With these points in mind,
studies were included if they:
1. Presented data regarding academic self-efficacy, measured in ac-
cordance with self-efficacy theory, at any level of specificity (see
Table 1).
2. Collected data regarding academic performance on a specific occa-
sion.
K. Talsma et al. Learning and Individual Differences 61 (2018) 136–150

Table 1
Examples of self-efficacy measures: specificity and inclusion/exclusion.

Excluded Included - task Included - domain Included - academic

“I am confident of my own
decision.” (Hwang et al.,
2016)
“I am a fast learner”
(Richardson et al., 2012)
“I am good at maths.” (Scott,
2000)
Confidence ratings for ability to achieve a
range of grades on an impending exam.
(Mone et al., 1995)
Confidence ratings for ability to correctly
answer specific questions on an impending
exam (Bonne & Johnston, 2016)
Confidence ratings for ability to perform
particular writing tasks (Meier, McCarthy, &
Schmeck, 1984)
“I'm sure I can do even the hardest work in my
math class.” (Lewis et al., 2012)
“I can explain the facts, concepts and
arguments covered in this course to others in
my own words” (Galyon et al., 2012)
Confidence ratings for achieving a range of
course grades (Beck & Schmidt, 2013)
“I'm confident I can master the courses I'm
taking this semester.” (Bong, 2001)
“I'm certain I can figure out how to do the
most difficult class work.” (Hornstra, Van Der
Veen, Peetsma, & Volman, 2013)
“How well can you pass all of your subjects in
school?” (Galla et al., 2014)

3. Collected data on self-efficacy and performance using a panel design
in accordance with the path model shown in Fig. 1, providing a set
of six correlations.
4. Included participants in traditional primary, secondary and tertiary
educational settings.
5. No interventions were conducted between the measurement waves,
as interventions may obscure the effects of self-efficacy and aca-
demic performance on each other. Control group data were eligible
for inclusion if other inclusion criteria were met.
We note that, in a single measurement wave, some researchers
measure self-efficacy immediately prior to performance (e.g., Galyon
et al., 2012), while some measure self-efficacy in the days or weeks
before the performance measurement (e.g., Richard, Diefendorff, &
Martin, 2006). Often, this is for practical reasons: to increase partici-
pation (Beck & Schmidt, 2013), or to reduce effects of exposure to ac-
tual test questions in a self-efficacy measure (Bonne & Johnston, 2016).
Research indicates that there is little difference in the prediction of
subsequent performance whether self-efficacy is measured immediately
prior or up to two weeks prior to performance (Beck & Schmidt, 2013;
Mone et al., 1995). Thus, studies in which self-efficacy is measured
within two weeks of performance were included, with degrees of con-
currency analysed as a risk of bias (Section 2.6).
2.2. Literature search
Electronic searches were conducted in Scopus, PsycINFO, the
Education Resources Information Centre (ERIC) and Web of Science.
Three search fields covered terms relating to self-efficacy, academic
performance (e.g., school, education, test), and methodological terms
(e.g., panel, longitudinal, lag). Full search parameters are available as
an online Supplementary file. In order to reduce the risk of publication
bias, unpublished papers such as dissertations and theses were eligible
for inclusion (Smith, 1980). In addition, there were no restrictions on
year of publication or language, as long as abstracts were available in
English.
A flow chart of the search is available as on online Supplementary
file. The final search was conducted in April 2016. In total, 5487 studies
were located in the database search, 3177 remained with duplicates
removed. After screening titles and abstracts for relevance, 197 articles
remained. We manually searched the reference lists of known extant
meta-analyses which included analyses of the relationship between self-
efficacy and academic performance. Where articles provided summaries
of included studies (Honicke & Broadbent, 2016; Robbins et al., 2004;
Sitzmann & Yeo, 2013; Valentine et al., 2004), these were used to
identify studies which measured both self-efficacy and performance and
had a longitudinal design, with 36 studies identified for potential in-
clusion. In other cases (Multon et al., 1991; Richardson et al., 2012), all
titles of included studies were screened for relevance (n = 252) and a
further 114 potentially relevant studies were identified. Full texts of
these 347 articles were obtained and screened according to the
inclusion criteria. If there was any uncertainty as to whether studies
met inclusion criteria, authors were contacted.
Twenty-three records were excluded because the required data were
not available after multiple attempts to contact the authors (about half
of these were dissertations). The vast majority of remaining exclusions
were made because studies did not measure self-efficacy or perfor-
mance (either at all, or in accordance with the specifications above) or
because studies did not collect data in a panel design. Some studies
superficially appeared to be panel designs (from their use of terms such
as “time 1” and “time 2” or “wave 1” and “wave 2”), but on consulta-
tion with authors, the variables in a given wave were actually measured
some months apart. Three studies measured self-efficacy subsequent to
performance at each measurement wave; these were excluded as this
approach is not consistent with recommendations of self-efficacy the-
orists to measure self-efficacy prior to performance (Moriarty, 2014).
Ultimately, 11 studies were identified which met the inclusion cri-
teria.
2.3. Data extraction
The first author used coding sheets to extract the required data from
included studies. Correlation matrices were taken from articles where
available (k = 4), or obtained from authors (k = 7). In order to ensure
independence of data points, and to avoid difficulties interpreting three
or more waves of data in panel analysis (Rogosa, 1988), each study
provided a single set of six correlations as per Fig. 1. Several studies
reported additional measurement waves; in order to fit data to the CLPA
model, only the first two waves were included. Several studies mea-
sured self-efficacy at multiple levels of specificity, in which case, the
measure which most closely matched the performance measure was
used, as such a match is believed to result in the greatest predictive
power (Klassen & Usher, 2010; Pajares, 1997). Coding also covered
sample size, publication status, study location, participant variables and
study characteristics relating to moderator analyses (Section 2.5) and
factors potentially associated with risk of bias (Section 2.6). A subset of
included studies was coded by KN and inter-rater agreement was 100%,
owing to the transparency of coding categories.
2.4. Overall analysis
Using Metafor for R (Viechtbauer, 2010), six pooled correlations
were calculated from the included studies' zero-order correlations. The
raw correlations were subjected to a Fisher's z transformation, and
random effects modelling was used as the included studies are believed
to represent a distribution of true effects (Borenstein, Hedges, &
Rothstein, 2007). Restricted maximum likelihood estimation was used
to ensure accurate estimation of variance (Thompson & Sharp, 1999),
and study weighting was based on the inverse of study-specific var-
iance, giving more weight to studies with greater precision
(Viechtbauer, 2010). Analyses were conducted on the zero-order cor-
relations using p-curve 4.05 (Simonsohn, Nelson, & Simmons, 2015) to

140
K. Talsma et al. Learning and Individual Differences 61 (2018) 136–150

calculate the underlying statistical power for the estimation of each of
the six pooled correlations. The matrix of pooled correlations resulting
from the above analysis served as input for the CLPA (Fig. 1) using
MPLUS 7 (Muthen & Muthen, 1998–2012).
2.5. Moderator analyses
The overall approach to moderator analyses was to test whether any
of the six pooled correlations forming the basis of the CLPA varied as a
function of the identified moderator variables. For moderators identi-
fied as significant, separate path models for subgroups of the moderator
variable were estimated using the same procedure as for the overall
analysis (Section 2.4). Subsequent to the estimation of the separate path
models for moderator subgroups, the cross-lagged path coefficients for
contrasting subgroups were tested for significant differences using the
pooled standard deviation method (Kock, 2014).
Tests of the effects of proposed moderators were undertaken using
meta-regression analyses in R using the metafor package (Viechtbauer,
2010). Within the framework of random-effects meta-analysis, meta-
regression is an analysis of the relationship between study-level vari-
ables and effect size – analogous to multiple regression in primary
studies assessing the relationship between subject-level variables and
an outcome (Borenstein, Hedges, Higgins, & Rothstein, 2009). As a
significant moderator effect for any individual pooled correlation could
result in changes to the overall meta-analytic path analysis, moderator
analyses were conducted for any moderator that showed significant Q
statistics for any of the six individual model paths. R2 values, which
indicate the amount of variance in the pooled correlations accounted
for by the moderator variable, were also calculated.
2.5.1. Moderator specification
Continuous values for moderators were available for sex (proportion
of males; k = 9), cohort (years since publication; the specific timing of
data collection was not published in most cases), and time lag between
waves (in weeks). Categorical moderator variables were necessary in
several cases as follows.
2.5.1.1. Age. Average participant age was not specified in more than
half of the studies included in this review, so a categorical moderator
(adults/children) was constructed using age data, where available, or
participant education level information (e.g., grade 5 versus tertiary
education). In this way, all studies could be included in the moderator
analysis.
scale was considered to be psychometrically less sound than a unipolar
scale (Bandura, 2012), this study was grouped with the Likert scales.
2.5.1.3. Specificity/match. Self-efficacy was coded as either domain-
based or task-based. In the included studies, all measures of
performance were incidentally task-based (e.g., exams, tests). This
rendered coding for both self-efficacy specificity and match between
self-efficacy and performance measures unnecessary, because these
groups overlapped completely. That is, because performance was
task-based, studies with task-based self-efficacy were characterised
both as being matched to performance and as having a more specific
level of self-efficacy, whereas studies with domain-based self-efficacy
measures were characterised both as being unmatched to performance
and as having a less specific level of self-efficacy. Therefore, only one
moderator analysis was required, and for ease of interpretation, we
refer hereafter to the moderator categories as “matched” or “non-
matched”.
2.6. Assessment of risk of bias
Risk of bias was assessed taking into account the reliability of self-
efficacy and performance measurements, the concurrency issue men-
tioned in Section 2.1, publication/selective reporting bias, and author
bias. We note that some potential sources of bias (e.g., atheoretical or
imprecise operationalisations of constructs) are minimised through the
application of the specific inclusion criteria outlined in Section 2.1.
2.6.1. Reliability of self-efficacy measurement
All reported Cronbach's alpha reliability coefficients for self-efficacy
measures were acceptable at > 0.70, with all but one exceeding
α = 0.83 (Nunnally & Bernstein, 1994). Nonetheless, self-efficacy re-
liability was analysed as a potential continuous moderator of the six
pooled correlations, using the same meta-regression approach as was
taken with the moderators (Section 2.5). None of the model paths were
significantly moderated by reliability (all Q statistics, p > 0.05);
therefore, reliability of the self-efficacy measure is not deemed to pre-
sent a risk of bias in the present study.
2.6.2. Reliability of performance measurement
Only three studies provided reliability information regarding per-
formance; therefore, a moderator analysis was not conducted. All per-
formance measures were taken from academic records and are there-
fore deemed to be at low risk of bias.

2.5.1.2. Scale. Unipolar scales were compared with Likert scales 2.6.3. Concurrency of measurement
categorically. In one case, the scale could not be identified as either As mentioned above, studies were defined as meeting the require-
unipolar or Likert (Vancouver & Kendall, 2006). As this single-item ments for panel measurement when self-efficacy and performance were

Table 2
Features of included studies including zero-order correlations and moderators.

Study n Age %m SE α CC Lag Match Scale SE1P1 SE1P2 SE2P1 SE2P2 SE1SE2 P1P2

Beck et al., 2013 69 A ns 0.94 No S No U 0.333 0.292 0.488 0.313 0.736 0.697
Bonne et al., 2016 50 C 36 ns No L Yes L 0.570 ns 0.370 0.310 0.300 0.390
Daniels et al., 2010 111 A ns 0.83 Yes S No U 0.201 0.229 0.429 0.219 0.676 0.541
Finney et al., 2003 103 21 32 0.91 Yes L Yes L 0.480 0.224 0.404 0.441 0.314 0.340
Galyon et al., 2012 213 A 26 0.92 Yes S Yes L 0.079 0.026 0.239 0.213 0.547 0.570
Lewis et al., 2012 1456 C 47 0.84 No L No L 0.318 0.272 0.285 0.353 0.399 0.680
Meier et al., 1984 71 A 57 0.85 Yes L Yes U 0.361 0.370 0.210 0.090 0.690 0.460
Mone, 1994 252 21.6 59 0.87 Yes S Yes U 0.310 0.290 0.480 0.320 0.670 0.490
Mone et al., 1995 215 A 53 0.85 No S Yes U 0.320 0.220 0.450 0.240 0.650 0.530
Richard et al., 2006 83 19.9 21 0.83 No S No L 0.340 0.340 0.480 0.390 0.760 0.630
Vancouver et al., 2006 65 21 21 0.83 No S Yes O 0.108 0.253 0.405 0.319 0.630 0.189

Note: mean age is provided where reported, otherwise, age group (adults = A, children = C) is provided based on sample information. SE α = Cronbach's alpha for self-efficacy; %
m = proportion males; CC = concurrency; L = long, S = Short; U = unipolar, L = Likert, O = Other; ns = not specified; SE1P1 = self-efficacy time 1 with performance time 1, etc.

141
K. Talsma et al. Learning and Individual Differences 61 (2018) 136–150

Table 3
Pooled correlations and tests of heterogeneity, publication bias and observed power: overall analysis.

k n SE1P1 SE1P2 SE2P1 SE2P2 SE1SE2 P1P2

Pooled correlation (SE) 11 2688 0.316⁎⁎⁎ (0.04) 0.248⁎⁎⁎ (0.03) 0.402⁎⁎⁎ (0.04) 0.312⁎⁎⁎ (0.03) 0.689⁎⁎⁎ (0.07) 0.583⁎⁎⁎ (0.06)
95% CI 0.231, 0.401 0.182, 0.313 0.328, 0.477 0.255, 0.369 0.551, 0.827 0.468, 0.697
I2 68.15 45.32 58.61 30.95 88.89 83.42
Q 26.007⁎⁎ 14.777 26.5777⁎⁎ 14.1855 191.9659⁎⁎⁎ 71.3827⁎⁎⁎
Failsafe N 749 431 1144 735 3144 2979
p-Curve, full (Z) − 9.56⁎⁎⁎ − 6.02⁎⁎⁎ − 13.31⁎⁎⁎ − 7.98⁎⁎⁎ − 19.99⁎⁎⁎ − 17.58⁎⁎⁎
p-Curve, half (Z) − 9.63⁎⁎⁎ − 5.26⁎⁎⁎ − 12.65⁎⁎⁎ − 7.28⁎⁎⁎ − 20.68⁎⁎⁎ − 17.08⁎⁎⁎
Observed power [90% CI] 99% [98%, 99%] 94% [77%, 99%] 99% [99%, 99%] 91% [91%, 99%] 99% [99%, 99%] 99% [99%, 99%]

Note: correlations are Fisher's Z transformations. SE1P1 = self-efficacy at time 1 with performance at time 1; SE1P2 = self-efficacy at time 1 with performance at time 2; SE2P1 = self-
efficacy at time 2 with performance at time 1; SE2P2 = self-efficacy at time 2 with performance at time 2; SE1SE2 = self-efficacy at time 1 with self-efficacy at time 2;
P1P2 = performance at time 1 with performance at time 2.
⁎⁎
p < 0.01.
⁎⁎⁎
p < 0.001.

measured within two weeks of each other at any given measurement
wave. In order to assess the validity of this approach to inclusion, a
categorical moderator variable was coded such that those studies in
which all measures were taken in the same sitting were compared with
all other studies. This variable did not significantly moderate any of the
pooled correlations that served as input for the path model (all Q sta-
tistics, p > 0.05). Therefore, differences in concurrency of measure-
ment are not deemed to present a risk of bias in the current study.
2.6.4. Selective publication/reporting bias
We screened both published and unpublished papers in the litera-
ture search process, and conducted both traditional (Failsafe N) and
contemporary (p-curve) analyses of the risk of selective publication/
reporting bias. Results (see Section 3) suggest minimal risk of bias based
on selective publication/reporting.
2.6.5. Author bias
Inclusion criteria and coding procedures were transparent and un-
ambiguous, resulting in 100% inter-rater agreement on coding in a
subset of studies; thus, the current meta-analysis is deemed to be at low
risk of author bias.
3. Results
Included studies (k = 11) were published between 1984 and 2016.
Study sample sizes ranged from 50 to 1456 (M = 244). Ten studies
were conducted in North America, with one study conducted in New
Zealand. There were ten published articles and one conference paper.
Ten studies comprised mostly white participants; one study comprised
mostly Hispanic participants. Six studies considered mathematical do-
mains, while the remainder considered management, writing, human
development, and psychology. Further details of included studies, in-
cluding zero-order correlations and moderator information, are in
Table 2.
3.1. Overall analysis
Table 3 shows the pooled estimates of the six correlations in the
cross-lagged panel model, along with standard errors, and 95% con-
fidence intervals. Also in Table 3 are the results of tests of hetero-
geneity, selective publication/reporting bias, and observed power. The
failsafe N values indicate that substantial numbers of additional null-
effect studies would be needed to increase the p-value for each path
to > 0.05 (Rosenthal, 1979). Data sets which are subject to selective
reporting or p-hacking are likely to show a left-skewed p distribution.
The full- and half- p-curve analyses all show right-skewed distributions,
indicating that the data included in the calculation of each of the six
model paths have evidential value (Simonsohn, Simmons, & Nelson,
2015).2 Observed power estimates, ranging from 91% to 99%, are also
reported in Table 3. The I2 and Q homogeneity test values suggest
sampling error alone is unlikely to account for the differences in effect
sizes. Forest and funnel plots are available as online Supplementary
material. Overall, the pooled correlations are consistent with previous
findings, with moderate-to-strong positive auto-correlations, and small-
to-moderate positive synchronous and cross-lagged correlations.
Fig. 2 displays the meta-analytic path model for the overall analysis,
estimated from the pooled correlations in Table 3. All of the paths are
positive and statistically significant, p < 0.001. Focusing on the cross-
lagged paths, the performance → self-efficacy effect was β = 0.205,
whereas the self-efficacy → performance effect was β = 0.071. These
estimates are based on a fully saturated model. A model was also tested
in which the cross-lagged paths were constrained to be equal, indicating
an equivalent reciprocal relationship. A Chi-squared test of the con-
strained model was interpreted as a Chi-squared difference test, com-
paring whether the model constraining the cross-lagged paths to be
equal resulted in a loss of fit compared to the model allowing them to
vary: if this test was significant, the constrained model provided a
significantly poorer fit than the saturated model (Menard, 2007). In this
case, the constrained model provided a significantly poorer fit to the
data, χ2 = 37.47, p < 0.0001; thus, the cross-lagged paths are sig-
nificantly different from each other.
Overall, these results indicate reciprocal effects: self-efficacy un-
iquely influences subsequent performance, and vice versa. Performance
is a significantly stronger antecedent of self-efficacy than the reverse:
the performance → self-efficacy effect is almost three times the size of
the self-efficacy → performance effect.
3.2. Moderator analyses
The relationship between academic self-efficacy and academic
performance was significantly moderated by sample age, length of the
time lag, the degree of match between the self-efficacy measure and the
performance measure, and the type of self-efficacy scale used. The
proportion of male participants and elapsed time since publication did
not moderate the relationship. Complete meta-regression analyses are
available in Supplementary material.
Path analyses were conducted on eight separate pooled correlation
matrices, with two subgroups for each of the four significant moderator
variables. For those variables which were already categorical, the
subgroups as specified in Section 2.5.1 were used for the separate path
2
The p-curve user guide directs authors to present a p-curve disclosure table. Such a
table is most applicable to syntheses of varied experimental data. In the present case,
much of the information that would be presented in such a table is identical across studies
(e.g., all designs and statistics are correlational). Other relevant information which would
form part of the disclosure table (e.g., statistical values) is in Table 2.

142
K. Talsma et al. Learning and Individual Differences 61 (2018) 136–150

Fig. 2. Cross-lagged panel model: overall analysis.

Note: figures are standardised path coefficients. A model in which the cross-lagged paths were constrained to be equal provided a significantly poorer fit to the data (see Section 3.1).
⁎⁎⁎
p < 0.001.

Table 4
Pooled correlations: moderator analyses.

n SE1P1 SE1P2 SE2P1 SE2P2 SE1SE2 P1P2

⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎ ⁎⁎⁎

Age group Adults 1182 r (SE) 0.290 (0.05) 0.240 (0.04) 0.427 (0.04) 0.288 (0.03) 0.758 (0.06) 0.562⁎⁎⁎ (0.06)
95% CI 0.196, 0.383 0.161, 0.320 0.345, 0.509 0.226, 0.351 0.631, 0.885 0.448, 0.676
Children 1506 r (SE) 0.456⁎⁎ (0.16) 0.279⁎⁎⁎ (0.02) 0.296⁎⁎⁎ (0.03) 0.367⁎⁎⁎ (0.03) 0.419⁎⁎⁎ (0.03) 0.645⁎⁎ (0.21)
95% CI 0.151, 0.761 0.228, 0.330 0.245, 0.347 0.317, 0.418 0.368, 0.470 0.239, 1.051
Lag Short 1008 r (SE) 0.249⁎⁎⁎ (0.05) 0.229⁎⁎⁎ (0.05) 0.448⁎⁎⁎ (0.05) 0.282⁎⁎⁎ (0.03) 0.797⁎⁎⁎ (0.05) 0.598⁎⁎⁎ (0.07)
95% CI 0.158, 0.341 0.137, 0.320 0.357, 0.539 0.220, 0.345 0.704, 0.889 0.470, 0.726
Long 1680 r (SE) 0.433⁎⁎⁎ (0.07) 0.281⁎⁎⁎ (0.02) 0.301⁎⁎⁎ (0.02) 0.333⁎⁎⁎ (0.07) 0.473⁎⁎⁎ (0.12) 0.544⁎⁎⁎ (0.12)
95% CI 0.295, 0.571 0.232, 0.329 0.253, 0.349 0.195, 0.470 0.241, 0.705 0.309, 0.778
Matched specificity Yes 969 r (SE) 0.328⁎⁎⁎ (0.07) 0.223⁎⁎⁎ (0.05) 0.397⁎⁎⁎ (0.05) 0.285⁎⁎⁎ (0.04) 0.643⁎⁎⁎ (0.08) 0.487⁎⁎⁎ (0.06)
95% CI 0.187, 0.469 0.120, 0.327 0.298, 0.495 0.210, 0.359 0.484, 0.802 0.377, 0.597
No 1719 r (SE) 0.323⁎⁎⁎ (0.02) 0.281⁎⁎⁎ (0.02) 0.420⁎⁎⁎ (0.07) 0.360⁎⁎⁎ (0.02) 0.779⁎⁎⁎ (0.14) 0.773⁎⁎⁎ (0.06)
95% CI 0.276, 0.371 0.233, 0.328 0.285, 0.554 0.312, 0.407 0.509, 1.05 0.664, 0.882
Scale Uni-polar 718 r (SE) 0.314⁎⁎⁎ (0.04) 0.275⁎⁎⁎ (0.04) 0.473⁎⁎⁎ (0.04) 0.265⁎⁎⁎ (0.04) 0.818⁎⁎⁎ (0.04) 0.592⁎⁎⁎ (0.04)
95% CI 0.240, 0.388 0.201, 0.349 0.399, 0.547 0.191, 0.339 0.744, 0.892 0.514, 0.671
Likert 1970 r (SE) 0.328⁎⁎⁎ (0.09) 0.221⁎⁎⁎ (0.06) 0.346⁎⁎⁎ (0.04) 0.349⁎⁎⁎ (0.04) 0.566⁎⁎⁎ (0.11) 0.546⁎⁎⁎ (0.10)
95% CI 0.160, 0.496 0.104, 0.338 0.261, 0.431 0.276, 0.422 0.358, 0.774 0.346, 0.747

Note: r = pooled correlations after Fisher's Z transformations; SE = self-efficacy; SE1P1 = self-efficacy at time 1 with performance at time 1; SE1P2 = self-efficacy at time 1 with
performance at time 2; SE2P1 = self-efficacy at time 2 with performance at time 1; SE2P2 = self-efficacy at time 2 with performance at time 2; SE1SE2 = self-efficacy at time 1 with self-
efficacy at time 2; P1P2 = performance at time 1 with performance at time 2.
⁎⁎
p < 0.01.
⁎⁎⁎
p < 0.001.

analyses (i.e., adults and children, matched and non-matched specifi-
city, and Likert and unipolar scales). Time lag had been analysed as a
continuous moderator; thus, we formed short and long categories for
comparison.3
Pooled correlations, standard errors and 95% confidence intervals
for the moderator categories are shown in Table 4, with a similar pat-
tern of correlations to the overall model. Homogeneity test values,
observed power, forest plots and funnel plots for pooled correlations for
moderator subgroups are available as online Supplementary material.
3
For this categorisation, we note that recent findings in the literature indicate that self-
efficacy is more strongly related to performance when measured up to 2 months prior to
the performance measure, with weaker relationships associated with longer lags (from
3 months upwards) (Gore Jr, 2006; Obrentz, 2012; Phan, 2012; Zusho et al., 2003). In the
included data set, this lag range coincided with a natural division in lag times: 7 studies
had lag times of up to 7 weeks, and 4 studies had longer lags of 3 to 12 months, while no
studies had a lag of between 7 weeks and 3 months. This categorisation therefore pro-
vided theoretically grounded as well as readily interpretable differentiation of studies
with a reasonable number of studies in each category.
Figs. 3 to 6 show the path models for the subgroups of moderator
variables found to impact significantly on the relationship between self-
efficacy and academic performance, estimated on the basis of the
pooled correlations in Table 4. These models were largely consistent in
that most paths were again positive and statistically significant. One
key exception is that, for children, the βSE1P2 cross-lagged path was
near-zero and did not reach statistical significance. In all cases, the
performance → self-efficacy paths were stronger than the corre-
sponding self-efficacy → performance paths.
As with the overall analyses, Chi-squared tests were conducted to
ascertain whether improved model-fit compared to saturated models
was obtained with cross-lagged paths constrained to be equal. With the
exception of long time lags (p = 0.07), models in which the cross-
lagged paths were constrained to be equal provided a significantly
poorer fit to the data (p < 0.001). Full details of these tests are
available as Supplementary material.
Table 5 shows the results of tests of differences between cross-
lagged path coefficients for moderator subgroups. The reciprocal

143
K. Talsma et al.
relationship was significantly stronger for adults than for children, for
both of the cross-lagged paths. Studies with shorter lags showed
stronger reciprocal effects than those with longer delays between
measurement waves; this difference was statistically significant for the
βP1SE2 path. Reciprocal effects were stronger for studies in which the
specificity of the self-efficacy measure matched the performance mea-
sure, though the differences did not reach statistical significance. Stu-
dies with unipolar self-efficacy scales showed stronger reciprocal effects
than those with Likert scales; this difference was statistically significant
for the βP1SE2 path.
4. Discussion
4.1. Overall analysis: I believe therefore I achieve, and vice versa
This study explored the chicken-and-egg conundrum in the re-
lationship between academic self-efficacy and academic performance
by means of a systematic review and meta-analytic CLPA. The meta-
analysis provided support for a reciprocal effects model: both self-effi-
cacy and performance had statistically significant, unique positive in-
fluences on each other over time (Fig. 2).
4.1.1. Interpretation
The finding of reciprocal effects is consistent with the conception of
reciprocal determinism in social cognitive theory (Bandura, 1977,
1997; Pajares & Usher, 2008). Reciprocity aligns with notions of the
self-efficacy ↔ performance relationship as a feedback loop, cycle or
spiral (Klassen & Usher, 2010; Lindsley et al., 1995; Multon et al., 1991;
Shea & Howell, 2000). The net performance → self-efficacy effect
highlights the important role of performance experience in the forma-
tion of self-efficacy beliefs (Britner & Pajares, 2006; Usher & Pajares,
2008). The finding of a unique positive self-efficacy → performance
influence while controlling for the opposite pathway in the CLPA casts
doubt on the claim that self-efficacy is merely reflective of past per-
formance (Bandura, 2012; Feltz et al., 2008; cf. Heggestad & Kanfer,
2005; Vancouver et al., 2001). Instead, the unique influence of self-
efficacy on performance highlights the generative or mobilising power
of self-efficacy, which reflects what you do with what you've got (Gist &
Mitchell, 1992). Consistent with this, a range of studies demonstrate
that self-efficacy affects goals, effort, resourcefulness, persistence, use
of cognitive strategies and choice of activities (see Bandura, 1997,
2012; Klassen & Usher, 2010; Pajares, 1996; Pajares & Schunk, 2001).
In gauging the practical significance of these results, we note that
the effect sizes for self-efficacy → performance are in the range pre-
viously found by Valentine et al. (2004) and Richardson et al. (2012).
144
Learning and Individual Differences 61 (2018) 136–150
The effect sizes for performance → self-efficacy are also similar in
magnitude. On first observation, it may be argued that these effects,
which, by convention are small-to-medium (Cohen, 1988), are unlikely
to have far-reaching implications. However, academic achievement
exists within a very intricate framework (Pajares, 2007), involving
factors both internal to the student; e.g., personality, coping styles
(Komarraju, Karau, Schmeck, & Avdic, 2011; MacCann, Fogarty,
Zeidner, & Roberts, 2011) and external to the student, e.g., parental
engagement, student-teacher relationships (Dotterer & Lowe, 2011;
Moreira, Dias, Vaz, & Vaz, 2013). In such a complex context, the con-
sistency and size of the present effects provide some evidence that they
are meaningful. The potency of the effect of self-efficacy on academic
performance may have been overstated in the literature because of
reliance on unidirectional findings. However, in the context of re-
ciprocal effects, there is cause for optimism that interventions targeting
either self-efficacy or performance will have flow-on effects to the other
variable – and interventions which target both constructs will have
synergistic effects. Conversely, considering either self-efficacy or per-
formance in a vacuum potentially means ignoring half of the equation.
4.1.2. Research implications
The finding of a reciprocal self-efficacy ↔ performance relationship
suggests that carefully timed measurement of both constructs will en-
hance research in this area. It is common in the literature to measure
self-efficacy on multiple occasions followed by a single performance
measurement, or to measure self-efficacy once, followed by staggered
measures of performance. Researchers also measure self-variables on
multiple occasions and then average the measurements. Such ap-
proaches are unlikely to provide accurate insights into how the two
variables relate over time. Of course, research must be designed ac-
cording to the specific research questions under consideration.
Generally, however, researchers should consider reciprocity when at-
tempting to isolate temporally dynamic features of the relationship
(Bandura & Jourden, 1991).
4.1.3. Practical implications
The finding of a reciprocal self-efficacy ↔ performance relationship
suggests that students would benefit when interventions to enhance
self-efficacy are combined with regular opportunities to experience
performance success. Approaches to enhancing academic self-efficacy
are detailed elsewhere (Pajares, 2006; Siegle & McCoach, 2007). We
provide several examples which are likely to have synergistic effects
owing to their anticipated concurrent impact on self-efficacy and per-
formance. Educators may observe that there is nothing novel here –
what is suggested is that these approaches will be of most benefit when
Fig. 3. Cross-lagged panel models for adults/chil-
dren respectively.
Note: figures are standardised path coefficients. For
both adults and children, a model in which the
cross-lagged paths were constrained to be equal
did not provide a significantly poorer fit to the data
(see online Supplementary material).
⁎⁎⁎
p < 0.001.
K. Talsma et al.
used in conjunction with each other in a cyclical fashion.
Guided mastery, which exists at the junction between self-efficacy
and performance, involves assisting students to navigate through in-
cremental authentic performance experiences, with scaffolding from
teachers being reduced over time (Bong, 2012; Pajares & Schunk,
2001). It is recommended that tasks are graded to individual learners so
that they are moderately challenging but accomplishable, and that
teachers help students set incremental goals that are specific and
proximal (Bong, 2012; Pajares, 2006). To provide evidence to students
of their growing mastery, progress toward goals should be made ex-
plicit, feedback should be timely and accurate, and success attributed to
internal, stable student characteristics/behaviors (Bong, 2012; Pajares,
2006; Siegle & McCoach, 2007).
An adaptive interpretation of performance experience may also
leverage reciprocal self-efficacy ↔ performance effects. Performance
should be framed in terms of progress rather than shortfalls, using in-
ternal rather than comparative standards, and showing how the ex-
perience can improve future performance (Bandura, 2012; Lindsley
et al., 1995; Pajares, 2006). Where performance is interpreted as a
failure, attributing this to factors under the student's control (e.g., in-
sufficient effort or ineffective strategy use) can provide a buffer for self-
efficacy beliefs (Bonne & Johnston, 2016; Pajares, 2006).
It must be noted that there is no single guaranteed way to enhance
self-efficacy beliefs or performance outcomes, and it has been suggested
that an assumption that it is always possible to generate lasting,
adaptive changes to self-efficacy beliefs is neither accurate nor helpful
(Gist & Mitchell, 1992; Klassen & Usher, 2010). Approaches should be
145
Learning and Individual Differences 61 (2018) 136–150
Fig. 4. Cross-lagged panel models for matched/non-
matched specificity respectively.
Note: figures are standardised path coefficients. For
both matched and non-matched specificity, a model in
which the cross-lagged paths were constrained to be
equal provided a significantly poorer fit to the data (see
online Supplementary material). ⁎⁎⁎p < 0.001,
⁎⁎
p < 0.01, ⁎p < 0.05.
explored with both contextual factors and the individual students in
mind (Klassen & Usher, 2010).
4.2. Moderator effects: When believing leads to achieving, and vice versa
4.2.1. Age
As anticipated, the reciprocal relationship between self-efficacy and
academic performance varied depending on participant age, with sig-
nificantly stronger cross lagged effects for adults than for children.
Importantly, the βSE1P2 cross-lagged path was not statistically sig-
nificant for children: self-efficacy did not have a net effect on sub-
sequent performance for children in this sample. That is, the relation-
ship between self-efficacy and academic performance appears to be
unidirectional for children, but reciprocal for adults. This suggests that,
in children, self-efficacy may be detached from the generative processes
generally ascribed to it, such as mobilisation of resources and strategies
which lead to better performance, which are argued to exist in adults.
This may be because the capacity to work with abstractions regarding
the self is cultivated as cognitive maturity develops (Harter, 1999). Self-
efficacy exists within a system of self-regulation, and individuals' self-
regulatory capacities are known to develop over time (Bandura, 1981;
Heckhausen & Dweck, 1998). Self-efficacy is influenced by a range of
sources of information, in conjunction with a cognitive appraisal of that
information (Bandura, 1997). It follows that, with greater cognitive
maturity and more educational experience, adults generate different
and, in particular, more precise appraisals of previous performance
outcomes or future task difficulty, with subsequent differences in self-
Fig. 5. Cross-lagged panel models for unipolar/Likert
self-efficacy scale respectively.
Note: figures are standardised path coefficients. For
both unipolar and Likert scales, a model in which the
cross-lagged paths were constrained to be equal pro-
vided a significantly poorer fit to the data (see online
Supplementary material). ⁎⁎⁎p < 0.001, ⁎⁎p < 0.01,
⁎
p < 0.05.
K. Talsma et al.
efficacy levels (Phan, 2012; Phan & Ngu, 2016; Zimmerman, 1989). The
differences between adults and children may also reflect a lack of ca-
libration between self-efficacy and performance for children, which is
also likely to develop with an accumulation of task experience and
feedback (Artino, 2012; Bandura, 2012).
Other differences between adults and children in each of the paths
forming the overall model may also contribute; for example, self-effi-
cacy appears to be more consistent over time for adults, which is con-
sistent with the notion that self-efficacy stabilises as individuals ap-
proach the middle life span (Oettingen & Gollwitzer, 2015). On the
other hand, performance is comparatively less stable over time for
adults than children, likely due to the tendency for performance tasks to
increase in difficulty more dramatically in adult educational settings
(Pokay & Blumenfeld, 1990). In the same vein, it is also possible that
age functions as a proxy for the type of academic tasks undertaken, with
simpler and more defined tasks in childhood, and more complex and
abstract tasks in adulthood.
One may be tempted to ascribe the lack of a significant finding for
the βSE1P2 path to a lack of power to detect the effect, given that there
are only two studies comprising children. However, there are several
indications that this is unlikely to be the case. When considering the
number of participants in each group, the sizes are roughly equal
(adults, n = 1182; children, n = 1506). In addition, the observed
power for the underlying pooled correlations is strong. Furthermore, we
note that all other paths in the model comprising children are statisti-
cally significant, suggesting that the lack of a significant βSE1P2 path is a
valid result.
These findings suggest that educators of adults should consider
leveraging the synergies inherent in the self-efficacy ↔ performance
relationship by using interventions which may enhance self-efficacy
directly, while also providing regular opportunities for students to
perform academic tasks successfully. Children may benefit from early
opportunities to learn about and engage in metacognitive functions
(Nietfeld, Cao, & Osborne, 2006), and to increase their understanding
of, and exposure to, performance-feedback cycles (Artino, 2012; Gist &
Mitchell, 1992; Gore Jr, 2006).
4.2.2. Methodological moderators: lag, specificity match, self-efficacy scale
type
Studies that used relatively short time lags between first and second
assessments (7 weeks or less) showed stronger cross-lagged relation-
ships than studies with longer lag times (3–12 months), as expected.
This is consistent with Bandura's (1997) position regarding temporal
disparity, which suggests that self-efficacy should be a more accurate
predictor if assessed more proximally to behaviour. In addition, it is
likely that a range of unmeasured variables influence the paths in the
146
Learning and Individual Differences 61 (2018) 136–150
Fig. 6. Cross-lagged panel models for short/long lag
respectively.
Note: figures are standardised path coefficients. For the
short lag, a model in which the cross-lagged paths were
constrained to be equal provided a significantly poorer
fit to the data (see online Supplementary material). For
the long lag, a model in which the cross-lagged paths
were constrained to be equal did not provide a sig-
nificantly poorer fit to the data (see online
Supplementary material). ⁎⁎⁎p < 0.001, ⁎⁎p < 0.01,
⁎
p < 0.05.
model; it follows that there would be more intervening variables over a
longer period of time, and also that these would impart a greater degree
of influence over that period (Valentine et al., 2004). Considering the
dearth of research regarding the impact of different time lags on the
relationship between self-efficacy and performance, it may be that the
longer time lags exceed the period during which the effects of interest
are at their maximum levels (Clegg, Jackson, & Wall, 1977). We note
that the reciprocal influences of self-efficacy and performance did not
differ significantly from each other in studies conducted over a longer
term. We speculate that over the course of repeated exposure to per-
formance-feedback cycles, the dual influences of self-efficacy and per-
formance may converge to a point of equilibrium.
Consistent with our hypothesis, effect sizes were larger when the
specificity of the self-efficacy and performance measures were matched
than when they were not matched. This is consistent with both the
theoretical position and recommendation for a micro-analytic research
strategy (Bandura, 1977, 2012) and with empirical evidence in this
regard (Bong, 2001; Pajares & Miller, 1995). Also as anticipated, effect
sizes were larger for studies which used unipolar scales than for those
which used Likert scales. This finding is consistent with commentary
above regarding the relative psychometric properties of different types
of scales when measuring self-efficacy (Bandura, 2006, 2012; Klassen &
Usher, 2010).
Overall, the moderator analyses found that measurement ap-
proaches congruent with those recommended by self-efficacy theorists
resulted in stronger effects. While these differences did not consistently
reach statistical significance, the self-efficacy → performance effect
sizes for those studies using theory-consistent approaches were con-
sistently approximately twice the size of those found in other studies. It
appears that criticisms of self-efficacy measurement are going unheard
by the research community, with negative ramifications for the quality
of self-efficacy research (Klassen & Usher, 2010). Especially as long-
itudinal research exploring self-efficacy continues to develop, with in-
creased need to detect more subtle changes over time, researchers are
urged to make well-informed decisions about the self-efficacy scales
used (Bandura, 2006; Bong, 2006; Lee & Bobko, 1994).
4.2.3. Sex and cohort effects
In this data set, sex was not identified as a significant moderator.
While Huang (2013) found a gender difference in self-efficacy in a
meta-analysis, the effect size was very small (g = 0.08), and was cal-
culated from 247 samples which were found to be highly hetero-
geneous, with individual effects ranging from g = −1.60 to g = 1.40.
Furthermore, findings of sex differences in self-beliefs and achievement
are not consistent across meta-analyses, with Hansford and Hattie
(1982) and Valentine et al. (2004) reporting no effect. It is possible that
K. Talsma et al.
Table 5
Tests of differences between cross-lagged path coefficients for moderator subgroups.
Path t p
Age group SE1P2 3.13 0.0018
P1SE2 3.71 0.0004
Lag SE1P2 0.78 0.4200
P1SE2 4.33 < 0.0001
Match SE1P2 1.19 0.2327
P1SE2 0.77 0.4435
Scale SE1P2 1.37 0.1717
P1SE2 1.77 0.0342
there was no such effect in the samples reviewed here, or that the
number of included studies provided insufficient power to detect a very
small effect.
Time since publication was not identified as a significant moderator,
suggesting that cohort effects are not responsible for the patterns of
variance observed in the included data.
4.3. Limitations
Given the narrow inclusion criteria of this study, a small number of
studies was uncovered. This may be considered a limitation of the
present analyses. However, we note that no formal minimum number of
studies exists in order for meta-analysis to be useful – results from as
few as two studies can be fruitfully synthesised (Valentine, Pigott, &
Rothstein, 2010). Meta-analysis may not be suitable for a small group of
studies showing wildly different effects. In this case, however, a range
of assessments of the nature of the included data provide support for the
validity of the present findings. For example, there are several indica-
tions that the correlation matrices used as input for the path analyses
are stable: consistent effects across studies are shown in forest plots
(online Supplementary material), failsafe N values are substantial
(Table 3), and there is strong underlying statistical power for the esti-
mation of each of the model paths (Table 3). Furthermore, the lower
bounds of the 95% confidence intervals for the pooled correlations not
only confirm the statistical significance of the effects, but also suggest
that the smallest possible estimates of effects based on the included data
would still be deemed meaningful in the social sciences (Cohen, 1988).
Nonetheless, the small number of studies does pose limitations in terms
of interpretation: in particular, when interpreting the moderator ana-
lysis regarding age, we note that the studies focusing on children con-
sidered only the mathematical domain, and therefore we caution that
this finding may not generalise to other areas of study.
Consistent with previous reports (Honicke & Broadbent, 2016;
Multon et al., 1991; Richardson et al., 2012; Robbins et al., 2004;
Valentine et al., 2004), significant heterogeneity across studies was
observed in the present review (Table 3). However, the I2 value of 45%
for the self-efficacy → performance path for the SE-first studies (the
path that is comparable to previous unidirectional meta-analyses)
compares favourably to the values of ~70–90% reported in the only
review which published comparable statistics (Richardson et al., 2012).
By convention, an I2 of 45% is indicative of moderate heterogeneity;
values of 70–90% indicate substantial-to-considerable heterogeneity
(Higgins & Green, 2008). Thus, we venture that the fine-grained ap-
proach taken has resulted in a more homogeneous group of studies for
analysis. Even so, we note that the strict methodological inclusion cri-
teria left little room to restrict the sample further based on other con-
textual factors, such as domain of study, and that included studies may
therefore vary in ways that were not addressed by our moderator
analyses.
A further limitation is that the pooled correlations in the path
analyses are subject to measurement error, whereas the path analysis
assumes no such error (Selig & Little, 2012). However, given the sta-
bility of the input data as described above, this issue is mitigated.
147
Learning and Individual Differences 61 (2018) 136–150
Within self-efficacy theory, it is an individual's interpretation of their
performance which constitutes mastery experience (Klassen & Usher,
2010; Usher & Pajares, 2008). A B grade is likely to be interpreted as a
success for a C-average student, but as a failure for a straight-A student
(Pajares, 2006). By focusing on objective performance, the inter-
pretative mechanism linking performance to self-efficacy may be ob-
scured in some cases.
With the search strategy we implemented, we were able to exclude
studies which called a construct self-efficacy but measured it in a way
inconsistent with social cognitive theory, but we were not able to in-
clude studies which measured a construct consistent with social cogni-
tive theory but called something else, rendering our search potentially
incomplete. However, an approach which would have ensured the in-
clusion of theoretically consistent but unlabelled self-efficacy studies
seems unfeasible, and researchers conducting studies within a social
cognitive framework seem likely to use theory-consistent terminology.
The present review, being based on non-experimental data, pre-
cludes firm conclusions about causality. In this case, the concern is that
the model is not fully specified, owing to the abundance of factors af-
fecting human behaviour (Rogosa, 1988; Selig & Little, 2012). A finding
of reciprocal effects may point to a blurring of the underlying processes
which lead to self-efficacy and then to academic performance, or vice
versa. Experiments conducted at a micro-analytic level are required
when the aim is to detect causality, and further experimental research
in academic settings is recommended. That being said, it has been noted
that self-beliefs do not readily lend themselves to experimental ma-
nipulation: successful manipulation is likely to concurrently modify
other, potentially unmeasured, mental processes (Valentine et al.,
2004). In this context, cross-lagged panel analysis is considered to be
“an important tool for building an argument for the causal effect of one
variable on another” (Selig & Little, 2012, p. 271). In an area where
complete experimental control is unlikely to be achieved, these types of
analyses are of considerable value.
4.4. Directions for future research
While all included studies contained measures of both self-efficacy
and performance on two occasions, in most cases (k = 7), full corre-
lation matrices were not published and were sought from the authors.
Therefore, not only are there very few studies that collect panel data on
these constructs, there are even fewer that leverage this data to explore
reciprocity and the strength of directional effects. The lack of available
information regarding SES of participants precluded a moderator ana-
lysis on that basis, and the lack of variability in terms of participant race
and location meant that we were unable to undertake moderator ana-
lyses for these variables. Most participants were white North-
Americans; this may affect the generalisability of our findings. These
issues point to the need for further longitudinal research generally, and
particularly with socially and culturally diverse participants.
Children were also under-represented in included studies. Given the
contrasting findings for adults and children, the literature would be
enhanced by an understanding of how the self-efficacy ↔ performance
relationship changes over the course of childhood and adolescence. It
appears that the relationship may change from unidirectional to re-
ciprocal over this period. In the present analysis, where a gross dis-
tinction had to be made between adults and children, it was not pos-
sible to explore this.
We note that additional waves of data were available in some in-
cluded studies, but could not be incorporated into the current analyses.
Further meta-analytic research using change modelling approaches is
warranted when sufficient primary studies become available.
4.5. Conclusion
In this systematic review with meta-analytic cross-lagged panel
analysis, a reciprocal relationship between self-efficacy and academic
K. Talsma et al.
performance was found, consistent with Bandura's view of self-efficacy
within a framework of reciprocal determinism (Bandura, 1977, 1997).
This suggests that interventions designed to influence either or both
constructs may be fruitfully applied. To make a financial analogy, in-
vestment in either self-efficacy or performance improvements is likely
to yield compound interest. In developing curricula, combining direct
enhancement of self-efficacy with regular opportunities for perfor-
mance success is likely to take best advantage of the identified re-
ciprocal effects.
The self-efficacy ↔ performance relationship varied as a function of
participant age, self-efficacy scale type, measurement lag and match of
specificity between measures. A reciprocal relationship emerged for
adults, but this did not hold for children. Across methodological mod-
erators, those studies with approaches consistent with self-efficacy
theory consistently showed stronger reciprocal effects.
In closing, we note that the effects reported here, while modest, are
embedded within a complex system involving various influences on
self-efficacy and academic performance in isolation, as well as the
subtleties of their interrelation over time. A deeper understanding of
the reciprocal nature of the relationship between the two, and the
comparative strength of directional effects, is likely to be of value to
both researchers and educational practitioners.
Appendix A. Supplementary data
Supplementary data to this article can be found online at https://
doi.org/10.1016/j.lindif.2017.11.015.
References4
Artino, A. R. (2012). Academic self-efficacy: From educational theory to instructional
practice. Perspectives on Medical Education, 1(2), 76–85. http://dx.doi.org/10.1007/
s40037-012-0012-5.
Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change.
Psychological Review, 84(2), 191.
Bandura, A. (1981). Self-referent thought: A developmental analysis of self-efficacy. Social
Cognitive Development: Frontiers and Possible Futures, 200–239.
Bandura, A. (1986). Social foundations of thought and action: A social cognitive theory.
Englewood Cliffs, NJ: Prentice-Hall, Inc.
Bandura, A. (1997). Self-efficacy: The exercise of control. Macmillan.
Bandura, A. (2006). Guide for constructing self-efficacy scales. Self-efficacy Beliefs of
Adolescents, 5, 307–337.
Bandura, A. (2012). On the functional properties of perceived self-efficacy revisited.
Journal of Management, 38(1), 9–44. http://dx.doi.org/10.1177/0149206311410606.
Bandura, A. (2015). On deconstructing commentaries regarding alternative theories of
self-regulation. Journal of Management, 41(4), 1025–1044.
Bandura, A., & Jourden, F. J. (1991). Self-regulatory mechanisms governing the impact of
social comparison on complex decision making. Journal of Personality and Social
Psychology, 60(6), 941.
Bandura, A., & Locke, E. A. (2003). Negative self-efficacy and goal effects revisited.
Journal of Applied Psychology, 88(1), 87–99. http://dx.doi.org/10.1037/0021-9010.
88.1.87.
*Beck, J. W., & Schmidt, A. M. (2013). State-level goal orientations as mediators of the
relationship between time pressure and performance: A longitudinal study. Journal of
Applied Psychology, 98(2), 354–363. http://dx.doi.org/10.1037/a0031145.
Bong, M. (2001). Role of self-efficacy and task-value in predicting college students' course
performance and future enrollment intentions. Contemporary Educational Psychology,
26(4), 553–570. http://dx.doi.org/10.1006/ceps.2000.1048.
Bong, M. (2006). Asking the right question: How confident are you that you could suc-
cessfully perform these tasks. In F. Pajares, & T. Urdan (Eds.). Self-efficacy beliefs of
adolescents (pp. 287–305). Greenwich, CT: Information Age Publishing.
Bong, M. (2012). Self-efficacy. International guide to student achievement. 64–66.
*Bonne, L., & Johnston, M. (2016). Students' beliefs about themselves as mathematics
learners. Thinking Skills and Creativity, 20, 17–28. http://dx.doi.org/10.1016/j.tsc.
2016.02.001.
Borenstein, M., Hedges, L., & Rothstein, H. (2007). Meta-analysis: Fixed effect vs. random
effects. Meta-analysis.com.
Borenstein, M., Hedges, L. V., Higgins, J. P. T., & Rothstein, H. R. (2009). Meta-regression.
Introduction to meta-analysis. John Wiley & Sons, Ltd.187–203.
Britner, S. L., & Pajares, F. (2006). Sources of science self-efficacy beliefs of middle school
students. Journal of Research in Science Teaching, 43(5), 485–499. http://dx.doi.org/
10.1002/tea.20131.
Brown, S. D., Tramayne, S., Hoxha, D., Telander, K., Fan, X., & Lent, R. W. (2008). Social
4
References marked with an asterisk were included in the meta-analysis.
148
Learning and Individual Differences 61 (2018) 136–150
cognitive predictors of college students' academic performance and persistence: A
meta-analytic path analysis. Journal of Vocational Behavior, 72(3), 298–308. http://
dx.doi.org/10.1016/j.jvb.2007.09.003.
Caprara, G. V., Vecchione, M., Alessandri, G., Gerbino, M., & Barbaranelli, C. (2011). The
contribution of personality traits and self-efficacy beliefs to academic achievement: A
longitudinal study. British Journal of Educational Psychology, 81(1), 78–96.
Chiang, Y. T., & Lin, S. S. J. (2014). The measurement structure, stability and mediating
effects of achievement goals in math with middle-school student data. Scandinavian
Journal of Educational Research, 58(5), 513–527. http://dx.doi.org/10.1080/
00313831.2012.732607.
Chin, D., & Kameoka, V. A. (2002). Psychosocial and contextual predictors of educational
and occupational self-efficacy among Hispanic inner-city adolescents. Hispanic
Journal of Behavioral Sciences, 24(4), 448–464.
Clegg, C., Jackson, P., & Wall, T. (1977). The potential of cross-lagged correlation analysis
in field research. Journal of Occupational Psychology, 50(3), 177–196.
Cohen, J. (1988). Statistical power analysis for the behavioral sciences. Hillsdale, NJ:
Lawrence Erlbaum Associates.
*Daniels, M. A., Kain, J. M., Gillespie, J. Z., & Schmidt, A. M. (2010). Latent profiles of goal
orientation and motivational outcomes. Paper presented at the 70th annual Academy of
Management meeting, Montreal, Canada.
Davis-Kean, P. E., Huesmann, L. R., Jager, J., Collins, W. A., Bates, J. E., & Lansford, J. E.
(2008). Changes in the relation of self-efficacy beliefs and behaviors across devel-
opment. Child Development, 79(5), 1257–1269. http://dx.doi.org/10.1111/j.1467-
8624.2008.01187.x.
Dotterer, A., & Lowe, K. (2011). Classroom context, school engagement, and academic
achievement in early adolescence. Journal of Youth and Adolescence, 40(12),
1649–1660. http://dx.doi.org/10.1007/s10964-011-9647-5.
Eccles, J. S., & Blumenfeld, P. (1985). Classroom experiences and student gender: Are
there differences and do they matter. Gender Influences in Classroom Interaction,
79–114.
Feltz, D. L. (1982). Path analysis of the causal elements in Bandura's theory of self-efficacy
and an anxiety-based model of avoidance behavior. Journal of Personality and Social
Psychology, 42(4), 764–781.
Feltz, D. L., Chow, G. M., & Hepler, T. J. (2008). Path analysis of self-efficacy and diving
performance revisited. Journal of Sport and Exercise Psychology, 30(3), 401.
*Finney, S. J., & Schraw, G. (2003). Self-efficacy beliefs in college statistics courses.
Contemporary Educational Psychology, 28(2), 161–186. http://dx.doi.org/10.1016/
S0361-476X(02)00015-2.
Flavell, J. H. (1979). Metacognition and cognitive monitoring: A new area of cogniti-
ve–developmental inquiry. American Psychologist, 34(10), 906.
Galla, B. M., Wood, J. J., Tsukayama, E., Har, K., Chiu, A. W., & Langer, D. A. (2014). A
longitudinal multilevel model analysis of the within-person and between-person ef-
fect of effortful engagement and academic self-efficacy on academic performance.
Journal of School Psychology, 52(3), 295–308. http://dx.doi.org/10.1016/j.jsp.2014.
04.001.
*Galyon, C. E., Blondin, C. A., Yaw, J. S., Nalls, M. L., & Williams, R. L. (2012). The
relationship of academic self-efficacy to class participation and exam performance.
Social Psychology of Education, 15(2), 233–249. http://dx.doi.org/10.1007/s11218-
011-9175-x.
Garriott, P. O., & Flores, L. Y. (2013). The role of social cognitive factors in Mexican
American students' educational goals and performance: A longitudinal analysis.
Journal of Latina/o Psychology, 1(2), 85–94. http://dx.doi.org/10.1037/a0032378.
Gist, M. E., & Mitchell, T. R. (1992). Self-efficacy: A theoretical analysis of its determi-
nants and malleability. Academy of Management Review, 17(2), 183–211.
Goldstein, H. (1983). Measuring changes in educational attainment over time: Problems
and possibilities. Journal of Educational Measurement, 20(4), 369–377.
Gore Jr, P. A. (2006). Academic self-efficacy as a predictor of college outcomes: Two
incremental validity studies. Journal of Career Assessment, 14(1), 92–115. http://dx.
doi.org/10.1177/1069072705281367.
Hansford, B. C., & Hattie, J. A. (1982). The relationship between self and achievement/
performance measures. Review of Educational Research, 52(1), 123–142. http://dx.
doi.org/10.3102/00346543052001123.
Harter, S. (1999). The construction of the self: A developmental perspective. Guilford Press.
Heckhausen, J., & Dweck, C. S. (1998). Motivation and self-regulation across the life span.
Cambridge University Press.
Heggestad, E. D., & Kanfer, R. (2005). The predictive validity of self-efficacy in training
performance: Little more than past performance. Journal of Experimental Psychology.
Applied, 11(2), 84–97. http://dx.doi.org/10.1037/1076-898X.11.2.84.
Higgins, J. P., & Green, S. (2008). Cochrane handbook for systematic reviews of interventions.
Vol. 5. Wiley Online Library.
Honicke, T., & Broadbent, J. (2016). The influence of academic self-efficacy on academic
performance: A systematic review. Educational Research Review, 17, 63–84. http://dx.
doi.org/10.1016/j.edurev.2015.11.002.
Hornstra, L., Van Der Veen, I., Peetsma, T., & Volman, M. (2013). Developments in
motivation and achievement during primary school: A longitudinal study on group-
specific differences. Learning and Individual Differences, 23(1), 195–204. http://dx.
doi.org/10.1016/j.lindif.2012.09.004.
Huang, C. (2013). Gender differences in academic self-efficacy: A meta-analysis. Vol. 28, 233
Spring Street, New York, NY 10013: Springer1–35.
Hwang, M. H., Choi, H. C., Lee, A., Culver, J. D., & Hutchison, B. (2016). The relationship
between self-efficacy and academic achievement: A 5-year panel analysis. Asia-Pacific
Education Researcher, 25(1), 89–98. http://dx.doi.org/10.1007/s40299-015-0236-3.
Klassen, R. M. (2004). A cross-cultural investigation of the efficacy beliefs of South Asian
immigrant and Anglo Canadian nonimmigrant early adolescents. Journal of
Educational Psychology, 96(4), 731–742. http://dx.doi.org/10.1037/0022-0663.96.4.
731.
K. Talsma et al.
Klassen, R. M., & Usher, E. L. (2010). Self-efficacy in educational settings: Recent research
and emerging directions. 16(Part A), 1–33.
Kock, N. (2014). Advanced mediating effects tests, multi-group analyses, and measure-
ment model assessments in PLS-based SEM. International Journal of e-Collaboration
(IJeC), 10(1), 1–13.
Komarraju, M., Karau, S. J., Schmeck, R. R., & Avdic, A. (2011). The big five personality
traits, learning styles, and academic achievement. Personality and Individual
Differences, 51(4), 472–477. http://dx.doi.org/10.1016/j.paid.2011.04.019.
Lee, C., & Bobko, P. (1994). Self-efficacy beliefs: Comparison of five measures. Journal of
Applied Psychology, 79(3), 364.
Lee, H. S., Flores, L. Y., Navarro, R. L., & Kanagui-Munoz, M. (2015). A longitudinal test of
social cognitive career theory's academic persistence model among Latino/a and
White men and women engineering students. Journal of Vocational Behavior, 88,
95–103. http://dx.doi.org/10.1016/j.jvb.2015.02.003.
Lent, R. W., Lopez, F. G., & Bieschke, K. J. (1991). Mathematics self-efficacy: Sources and
relation to science-based career choice. Journal of Counseling Psychology, 38(4),
424–430.
*Lewis, J. L., Ream, R. K., Bocian, K. M., Cardullo, R. A., Hammond, K. A., & Fast, L. A.
(2012). Con Cariño: Teacher caring, math self-efficacy, and math achievement among
Hispanic English learners. Teachers College Record, 114(7), 1–42.
Lindsley, D. H., Brass, D. J., & Thomas, J. B. (1995). Efficacy-performing spirals: A
multilevel perspective. Academy of Management Review, 20(3), 645–678. http://dx.
doi.org/10.5465/amr.1995.9508080333.
MacCann, C., Fogarty, G. J., Zeidner, M., & Roberts, R. D. (2011). Coping mediates the
relationship between emotional intelligence (EI) and academic achievement.
Contemporary Educational Psychology, 36(1), 60–70. http://dx.doi.org/10.1016/j.
cedpsych.2010.11.002.
Majer, J. M. (2009). Self-efficacy and academic success among ethnically diverse first-
generation community college students. Journal of Diversity in Higher Education, 2(4),
243–250. http://dx.doi.org/10.1037/a0017852.
Matsui, T., Matsui, K., & Ohnishi, R. (1990). Mechanisms underlying math self-efficacy
learning of college students. Journal of Vocational Behavior, 37(2), 225–238. http://
dx.doi.org/10.1016/0001-8791(90)90042-Z.
*Meier, S., McCarthy, P. R., & Schmeck, R. R. (1984). Validity of self-efficacy as a pre-
dictor of writing performance. Cognitive Therapy and Research, 8(2), 107–120. http://
dx.doi.org/10.1007/BF01173038.
Menard, S. (2007). Handbook of longitudinal research: Design, measurement, and analysis.
Elsevier.
Mitchell, T. R., Hopper, H., Daniels, D., George-Falvy, J., & James, L. R. (1994).
Predicting self-efficacy and performance during skill acquisition. Journal of Applied
Psychology, 79(4), 506–517.
Moher, D., Liberati, A., Tetzlaff, J., & Altman, D. G. (2009). Preferred reporting items for
systematic reviews and meta-analyses: The PRISMA statement. Annals of Internal
Medicine, 151(4), 264–269.
*Mone, M. A. (1994). Comparative validity of two measures of self-efficacy in predicting
academic goals and performance. Educational and Psychological Measurement, 54(2),
516–529. http://dx.doi.org/10.1177/0013164494054002026.
*Mone, M. A., Baker, D. D., & Jeffries, F. (1995). Predictive validity and time dependency
of self-efficacy, self-esteem, personal goals, and academic performance. Educational
and Psychological Measurement, 55(5), 716–727. http://dx.doi.org/10.1177/
0013164495055005002.
Moreira, P. A. S., Dias, P., Vaz, F. M., & Vaz, J. M. (2013). Predictors of academic per-
formance and school engagement — Integrating persistence, motivation and study
skills perspectives using person-centered and variable-centered approaches. Learning
and Individual Differences, 24, 117–125. http://dx.doi.org/10.1016/j.lindif.2012.10.
016.
Moriarty, B. (2014). Research design and the predictive power of measures of self-effi-
cacy. Issues in Educational Research, 24(1), 55–66.
Multon, K. D., Brown, S. D., & Lent, R. W. (1991). Relation of self-efficacy beliefs to
academic outcomes: A meta-analytic investigation. Journal of Counseling Psychology,
38(1), 30–38.
Muthen, L. K., & Muthen, B. O. (1998-2012). Mplus user's guide (Seventh Edition). Los
Angeles, CA: Muthen & Muthen.
Nietfeld, J. L., Cao, L., & Osborne, J. W. (2006). The effect of distributed monitoring
exercises and feedback on performance, monitoring accuracy, and self-efficacy.
Metacognition and Learning, 1(2), 159–179.
Nunnally, J. C., & Bernstein, I. H. (1994). The assessment of reliability. In I. H. Bernstein,
& J. C. Nunnally (Vol. Eds.), Psychometric theory. Vol. 3. Psychometric theory (pp. 248–
292).
Obrentz, S. B. (2012). Predictors of science success: The impact of motivation and learning
strategies on college chemistry performance.
Oettingen, G., & Gollwitzer, P. M. (2015). Self-regulation in adolescence. Cambridge
University Press.
Pajares, F. (1996). Self-efficacy beliefs in academic settings. Review of Educational
Research, 66(4), 543–578.
Pajares, F. (1997). Current directions in self-efficacy research. Advances in Motivation and
Achievement, 10(149).
Pajares, F. (2006). Self-efficacy during childhood and adolescence: Implications for tea-
chers and parents. In F. Pajares, & T. Urdan (Eds.). Self-efficacy beliefs of adolescents
(pp. 339–368). Greenwich, CT: Information Age.
Pajares, F. (2007). Culturalizing educational psychology. In F. Salili, & R. Hoosain (Eds.).
Culture, motivation, and learning: A multicultural perspective (pp. 19–42). Charlotte, NC:
IAP.
Pajares, F., Hartley, J., & Valiante, G. (2001). Response format in writing self-efficacy
assessment: Greater discrimination increases prediction. Measurement and Evaluation
in Counseling and Development, 33(4), 214.
149
Learning and Individual Differences 61 (2018) 136–150
Pajares, F., & Kranzler, J. (1995). Self-efficacy beliefs and general mental ability in
mathematical problem-solving. Contemporary Educational Psychology, 20(4),
426–443.
Pajares, F., & Miller, M. D. (1995). Mathematics self-efficacy and mathematics perfor-
mances: The need for specificity of assessment. Journal of Counseling Psychology,
42(2), 190–198.
Pajares, F., & Schunk, D. H. (2001). Self-beliefs and school success: Self-efficacy, self-
concept, and school achievement. In R. Riding, & S. Rayner (Eds.). Perception (pp.
239–266). London: Ablex Publishing.
Pajares, F., & Usher, E. L. (2008). Self-efficacy, motivation, and achievement in school
from the perspective of reciprocal determinism. Advances in Motivation and
Achievement, 15, 391–423.
Parker, P. D., Marsh, H. W., Ciarrochi, J., Marshall, S., & Abduljabbar, A. S. (2014).
Juxtaposing math self-efficacy and self-concept as predictors of long-term achieve-
ment outcomes. Educational Psychology, 34(1), 29–48. http://dx.doi.org/10.1080/
01443410.2013.797339.
Phan, H. P. (2012). Informational sources, self-efficacy and achievement: A temporally
displaced approach. Educational Psychology, 32(6), 699–726. http://dx.doi.org/10.
1080/01443410.2012.708320.
Phan, H. P., & Ngu, B. H. (2016). Sources of self-efficacy in academic contexts: A long-
itudinal perspective. School Psychology Quarterly. http://dx.doi.org/10.1037/
spq0000151.
Pokay, P., & Blumenfeld, P. C. (1990). Predicting achievement early and late in the se-
mester: The role of motivation and use of learning strategies. Journal of Educational
Psychology, 82(1), 41–50. http://dx.doi.org/10.1037/0022-0663.82.1.41.
*Richard, E. M., Diefendorff, J. M., & Martin, J. H. (2006). Revisiting the within-person
self-efficacy and performance relation. Human Performance, 19(1), 67–87. http://dx.
doi.org/10.1207/s15327043hup1901_4.
Richardson, M., Abraham, C., & Bond, R. (2012). Psychological correlates of university
students' academic performance: A systematic review and meta-analysis.
Psychological Bulletin, 138(2), 353–387.
Riketta, M. (2008). The causal relation between job attitudes and performance: A meta-
analysis of panel studies. Journal of Applied Psychology, 93(2), 472.
Robbins, S. B., Lauver, K., Le, H., Davis, D., Langley, R., & Carlstrom, A. (2004). Do
psychosocial and study skill factors predict college outcomes? A meta-analysis.
Psychological Bulletin, 130(2), 261–288. http://dx.doi.org/10.1037/0033-2909.130.
2.261.
Rogosa, D. (1988). Analysis of reciprocal effects. In J. P. Keeves (Ed.). Educational re-
search, methodology and measurement: An international handbook. Elsevier Science &
Technology Books.
Rosenthal, R. (1979). The file drawer problem and tolerance for null results. Psychological
Bulletin, 86(3), 638.
Schwarzer, R., & McAuley, E. (2016). The world is confounded: A comment on Williams
and Rhodes (2016). Health Psychology Review, 10(2), 133–135. http://dx.doi.org/10.
1080/17437199.2016.1162667.
Scott, L. A. (2000). A matter of confidence? A new (old) perspective on sex differences in
mathematics achievement. (AAI9955419). PsycINFO database.
Selig, J. P., & Little, T. P. (2012). Autoregressive and cross-lagged panel analysis for long-
itudinal data. Handbook of developmental research methods. 265.
Shea, C. M., & Howell, J. M. (2000). Efficacy-performance spirals: An empirical test.
Journal of Management, 26(4), 791–812. http://dx.doi.org/10.1177/
014920630002600409.
Siegle, D., & McCoach, D. B. (2007). Increasing student mathematics self-efficacy through
teacher training. Journal of Advanced Academics, 18(2), 278–312.
Simonsohn, U., Nelson, L. D., & Simmons, J. P. (2015). Official user-guide to the P-curve.
Retrieved from www.p-curve.com/guide.pdf.
Simonsohn, U., Simmons, J. P., & Nelson, L. D. (2015). Better P-curves: Making p-curve
analysis more robust to errors, fraud, and ambitious p-hacking, a reply to Ulrich and
Miller (2015). Journal of Experimental Psychology: General, 144(6), 1146–1152.
http://dx.doi.org/10.1037/xge0000104.
Singh, B., & Tyagi, T. K. (2014). The application of cross-lagged panel analysis in edu-
cational research. FACTA UNIVERSITATIS-Series Philosophy, Sociology, Psychology and
History, 2, 39–51.
Sitzmann, T., & Yeo, G. (2013). A meta-analytic investigation of the within-person self-
efficacy domain: Is self-efficacy a product of past performance or a driver of future
performance? Personnel Psychology, 66(3), 531–568.
Smith, M. L. (1980). Publication bias and meta-analysis. Evaluation in Education, 4, 22–24.
http://dx.doi.org/10.1016/0191-765X(80)90004-X.
Stankov, L., & Lee, J. (2014). Quest for the best non-cognitive predictor of academic
achievement. Educational Psychology, 34(1), 1–8. http://dx.doi.org/10.1080/
01443410.2013.858908.
Thompson, S. G., & Sharp, S. J. (1999). Explaining heterogeneity in meta-analysis: A
comparison of methods. Statistics in Medicine, 18(20), 2693–2708.
Usher, E. L., & Pajares, F. (2008). Sources of self-efficacy in school: Critical review of the
literature and future directions. Review of Educational Research, 78(4), 751–796.
Valentine, J. C., DuBois, D. L., & Cooper, H. (2004). The relation between self-beliefs and
academic achievement: A meta-analytic review. Educational Psychologist, 39(2),
111–133.
Valentine, J. C., Pigott, T. D., & Rothstein, H. R. (2010). How many studies do you need?:
A primer on statistical power for meta-analysis. Journal of Educational and Behavioral
Statistics, 35(2), 215–247. http://dx.doi.org/10.3102/1076998609346961.
*Vancouver, J. B., & Kendall, L. N. (2006). When self-efficacy negatively relates to mo-
tivation and performance in a learning context. Journal of Applied Psychology, 91(5),
1146–1153. http://dx.doi.org/10.1037/0021-9010.91.5.1146.
Vancouver, J. B., Thompson, C. M., & Williams, A. A. (2001). The changing signs in the
relationships among self-efficacy, personal goals, and performance. Journal of Applied
K. Talsma et al.
Psychology, 86(4), 605.
Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package.
Journal of Statistical Software, 36(3), 48. http://dx.doi.org/10.18637/jss.v036.i03.
Williams, T., & Williams, K. (2010). Self-efficacy and performance in mathematics:
Reciprocal determinism in 33 nations. Journal of Educational Psychology, 102(2),
453–466.
Zimmerman, B. J. (1989). A social cognitive view of self-regulated academic learning.
Journal of Educational Psychology, 81(3), 329.
Zimmerman, B. J. (2000). Self-efficacy: An essential motive to learn. Contemporary
150
View publication stats
Learning and Individual Differences 61 (2018) 136–150
Educational Psychology, 25(1), 82–91. http://dx.doi.org/10.1006/ceps.1999.1016.
Zimmerman, B. J., & Bandura, A. (1995). Self-efficacy and educational development. Self-
Efficacy in Changing Societies, 202–231.
Zimmerman, B. J., Bandura, A., & Martinez-Pons, M. (1992). Self-motivation for academic
attainment: The role of self-efficacy beliefs and personal goal setting. American
Educational Research Journal, 29(3), 663–676. http://dx.doi.org/10.2307/1163261.
Zusho, A., Pintrich, P. R., & Coppola, B. (2003). Skill and will: The role of motivation and
cognition in the learning of college chemistry. International Journal of Science
Education, 25(9), 1081–1094.